Model Hibrida HMM-XGBoost untuk Prediksi Sinyal Trading Cryptocurrency Berbasis Feature Augmentation Rezim Pasar

Fatih, Mohammad Faris Al (2026) Model Hibrida HMM-XGBoost untuk Prediksi Sinyal Trading Cryptocurrency Berbasis Feature Augmentation Rezim Pasar. Undergraduate thesis, UPN Veteran Jawa Timur.

[img] Text (COVER)
COVER.pdf

Download (1MB)
[img] Text (BAB I)
BAB I.pdf

Download (283kB)
[img] Text (BAB II)
BAB II.pdf
Restricted to Repository staff only until 9 November 2028.

Download (1MB)
[img] Text (BAB III)
BAB III.pdf
Restricted to Repository staff only until 9 November 2028.

Download (1MB)
[img] Text (BAB IV)
BAB IV.pdf
Restricted to Repository staff only until 9 November 2028.

Download (3MB)
[img] Text (BAB V)
BAB V.pdf

Download (274kB)
[img] Text (DAFTAR PUSTAKA)
DAFTAR PUSTAKA.pdf

Download (224kB)

Abstract

The cryptocurrency market is highly volatile, and technical indicators often give false signals when prices move sideways. This study tests whether market regime probabilities from a Hidden Markov Model (HMM), added as input features (feature augmentation), improve the performance of XGBoost in predicting Buy, Hold, and Sell signals. The data are 4-hour OHLC prices of SOL/USDT and XRP/USDT from Tokocrypto (January 1, 2021 to June 30, 2026), comprising 11,985 rows split chronologically into 80% training data and 20% test data. Scenario A (baseline) uses 17 technical indicators, while Scenario B (hybrid) adds three HMM regime probabilities (bullish, sideways, bearish) computed with the forward algorithm to prevent lookahead bias. Labels are built from an Average True Range (ATR) threshold, class weighting uses α = 0.5, and Grid Search evaluates 162 hyperparameter combinations. Scenario B raises precision from 0.3723 to 0.3944 and F1-score from 0.3542 to 0.3637 on SOL, and precision from 0.3726 to 0.4106 and F1-score from 0.3635 to 0.3800 on XRP. Feature importance analysis ranks RSI as the dominant feature, while the HMM regime features sit between ranks 9 and 19. In backtesting with a transaction cost of 0.4044%, Scenario B reduces losses: ROI improves from -6.00% to -4.62% on SOL and from -13.18% to -3.72% on XRP. Both models still record negative ROI and outperform only Buy and Hold, which loses -58.00% (SOL) and -54.97% (XRP).

Item Type: Thesis (Undergraduate)
Contributors:
ContributionContributorsNIDN/NIDKEmail
Thesis advisorParlika, RizkyNIDN0718058401rizkyparlika.if@upnjatim.ac.id
Thesis advisorMaulana, HendraNIDN1423128301hendra.maulana.if@upnjatim.ac.id
Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Q Science > QA Mathematics > QA76.6 Computer Programming
Q Science > QA Mathematics > QA76.87 Neural computers
Divisions: Faculty of Computer Science > Departemen of Informatics
Depositing User: Mohammad F Mohammad Faris Al Fatih
Date Deposited: 17 Nov 2026 04:09
Last Modified: 17 Nov 2026 04:09
URI: https://repository.upnjatim.ac.id/id/eprint/61273

Actions (login required)

View Item View Item